Sovereign home OS for AI cloud, coding, gaming, creative, and family entertainment. Base: Garuda Linux (Arch-based, Zen kernel, BTRFS+Snapper) Model-Agnostic: Supports any HuggingFace model.

Version: v16.5.0 | Base: Garuda Linux | License: MIT

Based on 4 independent code reviews of v16.4, v16.5 fixes 10 issues (3 P0 + 7 P1):

1. ✅ set -euo pipefail + tee pipeline bug (1-a) — QEMU/mkarchiso exit codes are now captured via PIPESTATUS[0] with set +e around tee pipelines (was being lost to tee's exit code)
2. ✅ OVMF fallback removed (2-a) — qemu-boot.sh now FAILS if OVMF_VARS.fd template is missing instead of silently copying from OVMF_CODE.fd (which produces broken UEFI NVRAM)
3. ✅ Deterministic QEMU serial capture (3-a) — console=tty0 console=ttyS0,115200n8 added to both archiso-x86_64.conf (UEFI) and syslinux.cfg (BIOS) — kernel now reliably outputs to serial port

4. ✅ README_VERSION validation (4-a) — build-iso.sh now validates VERSION against README.md in addition to profiledef.sh (build fails on any mismatch)
5. ✅ linux-lts kernel (5-a) — Added linux-lts + linux-lts-headers to packages.x86_64 as fallback when NVIDIA driver breaks on Zen kernel
6. ✅ model-manager.sh (6-b) — New weekly model updater script (/usr/local/bin/model-manager.sh) + systemd timer (Mondays 04:00). Full scope: pulls HuggingFace GGUF + Ollama tags, captures VRAM before/after, sends Telegram summary
7. ✅ downgrade AUR tool (7-a) — Added to packages.x86_64 for NVIDIA driver rollback (sudo downgrade nvidia-dkms)
8. ✅ vram subcommand (8-b) — Integrated into scarlix-mode (no separate script). Run scarlix-mode vram for per-GPU VRAM bars + health check (returns exit 2 if ≥90% utilization)
9. ✅ ZRAM tuned (9-a) — New /etc/systemd/zram-generator.conf: zstd compression + 32GB allocation (caps at RAM size on smaller systems)
10. ✅ BTRFS CoW disabled (10-a) — first-boot.sh now runs chattr +C on /models, /mnt/games, /var/lib/docker, /var/lib/scarlix immediately after mkdir (prevents fragmentation on large mutable stores)

• archiso profile structure (garuda-scarlix/)
• NVIDIA + CUDA installed post-install (keeps ISO small)
• Model-agnostic models.yaml with hf_repo/hf_file fields
• 6 GPU modes (ai/game/creative/turbo/offline/tv)
• 3-tier inference (SGLang → Ollama → llama.cpp)
• First-boot logging to /var/log/scarlix/first-boot.log

git clone https://github.com/MoZoHuJa/OS.git
cd OS
sudo pacman -S archiso edk2-ovmf qemu-desktop
bash installer/scripts/build-iso.sh

bash tests/qemu-boot.sh output/scarlix-os-v16.5-x86_64.iso

1. Write ISO to USB (Ventoy / dd)
2. Boot from USB → Calamares → auto BTRFS → auto user
3. TUI wizard: select "Main PC" or "HP Agent"
4. Wizard enables first-boot service + model-manager timer
5. Reboot → NVIDIA install + Docker services + model download (check /var/log/scarlix/first-boot.log)
6. Dashboard: http://192.168.1.100:8090

scarlix-mode {ai|game|creative|turbo|offline|tv|status|vram}

Edit /etc/scarlix/models.yaml → run download-models.sh → restart. Supports: Llama, Qwen, Mistral, Gemma, Phi, DeepSeek, Nemotron. Uses hf_repo field for direct GGUF downloads from HuggingFace. Weekly auto-update via model-manager.sh (Monday 04:00 + Telegram summary).

If NVIDIA driver breaks on Zen kernel:

# Option 1: rollback to previous driver version
sudo downgrade nvidia-dkms

# Option 2: boot LTS kernel
sudo grub-set-default 1   # select linux-lts entry
sudo reboot

• Algorithm: zstd (3.5x typical compression)
• Size: 32GB virtual (caps at RAM size on smaller systems)
• Config: /etc/systemd/zram-generator.conf
• Result: ~3x effective RAM for mixed AI + Docker + gaming workload

garuda-scarlix/
├── profiledef.sh                          # archiso profile (v16.5.0)
├── packages.x86_64                        # + linux-lts, downgrade, zram-generator
├── airootfs/
│   ├── etc/
│   │   ├── systemd/
│   │   │   ├── zram-generator.conf        # NEW: zstd + 32GB ZRAM (9-a)
│   │   │   └── system/
│   │   │       ├── first-boot.sh          # + chattr +C BTRFS CoW disable (10-a)
│   │   │       ├── model-manager.service  # NEW: weekly model updater (6-b)
│   │   │       ├── model-manager.timer    # NEW: Monday 04:00 trigger (6-b)
│   │   │       └── scarlix-first-boot.service
│   │   └── ...
│   └── usr/local/bin/
│       ├── scarlix-wizard                 # + enable model-manager.timer
│       ├── scarlix-mode                   # + `vram` subcommand (8-b)
│       └── model-manager.sh               # NEW: HF + Ollama + VRAM + Telegram (6-b)
├── efiboot/loader/entries/archiso-x86_64.conf   # + console=ttyS0 (3-a)
└── syslinux/syslinux.cfg                        # + console=ttyS0 (3-a)

installer/scripts/build-iso.sh             # + PIPESTATUS + README_VERSION check (1-a, 4-a)
tests/qemu-boot.sh                        # + PIPESTATUS + OVMF no-fallback (1-a, 2-a)
models.yaml                               # v16.5 comment update